Using Bayesian optimization to automate the calibration of complex hydrological models: Framework and application. (January 2022)

• Record Type: Journal Article
• Authors: Ma, Jinfeng; Zhang, Jing; Li, Ruonan; Zheng, Hua; Li, Weifeng
• Is Part Of: Environmental modelling & software. Volume 147(2022)
• Publication Date: 2022-01
• Subjects: Bayesian optimization -- Calibration -- Sensitivity analysis -- SWAT; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2021.105235
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: A framework that integrates Bayesian optimization (BO) and high-performance computing was developed, to automate calibration of complex hydrological models. It adopts a loosely coupled web architecture, integrating Tornado and SpringBoot, to facilitate bidirectional transfer of variables between BO and model evaluation. Extensive model evaluations were implemented on a Hadoop cluster, to wrap the model into the calculation flexibly and separate the calculation process from the algorithm execution effectively. A case study, calibrating a SWAT model in the Meichuan Basin (Jiangxi Province, China), indicated that the framework provides an ideal environment for assessment of the capability of BO to quantify the efficient estimation of SWAT parameters. Compared with that of the built-in SWAT-CUP tool, the number of executions was reduced from 1500 to 150, while maintaining similar accuracy. The framework also allows evaluation of the performance of different surrogate models and acquisition functions and provides instant visualization for searching for optimal parameters. Highlights: Automatic model calibration framework for Bayesian optimization (BO) was developed. Tornado and SpringBoot were integrated to enable transfer of variables between them. Hadoop cluster wrapped complex model internally and performed model evaluations simultaneously. Capability of BO to quantify efficient estimation of SWAT parameters was assessed. Performance of surrogate models and acquisition functions of BO was evaluated.